Separate calculations reproduced higher MMSE test error with a fixed EEG ratio in 59 ds004504 version 1.0.9 dementia participants

Separate calculations reproduced higher MMSE prediction error after adding one fixed theta to alpha EEG power ratio. Both studies used the same feature archive from OpenNeuro ds004504 version 1.0.9. The main sample had 36 people with Alzheimer disease and 23 with frontotemporal dementia. The baseline model used diagnosis, age, and recorded gender. Each test person was excluded from model fitting. Adding EEG increased the observed root mean squared error from 4.029 to 4.115 MMSE points. All 59 test predictions matched the concurrent publication within 6.04e-14 MMSE points. The adjusted association was 0.199 MMSE points per doubling of the geometric mean power ratio. Its approximate 95 percent HC3 interval was -2.218 to 2.616 points. This interval does not rule out moderate associations. This work adds 10,000 participant bootstrap samples, 9,999 residual permutations, and comparisons between recording periods. It also checks three source recordings and documents an unresolved difference in the source FTD score spread. The package preserves separate code, input features, scores, source hashes, results, figures, and retrieval instructions. This is a computational replication on shared inputs, with no new clinical sample or external test cohort.

electroencephalography · theta to alpha ratio · MMSE · Alzheimer disease · frontotemporal dementia · computational replication · leave-one-out prediction